[openib-general] CMA: compliancy issue?

[Or Gerlitz]

Michael S. Tsirkin wrote:
> Quoting r. Or Gerlitz <ogerlitz at voltaire.com>:
>> If the ULP has created a QP via the CMA then the CMA will send RTU and 
>> deliver up ESTABLISHED event, else the CMA will deliver up 
>> CONNECT_RESPONSE event and only later when the ULP calls rdma_accept the 
>> CMA will send the RTU.

> Correct, that's what I am saying.

> I want to create a QP via CMA (no sense in duplicating functionality), but I
> need CMA to sent RTU *after* delivering ESTABLISHED event, not before as it does
> now, and I need a way to tell CMA to reject the connection after my handler
> looks at private data.

Please note that if the ULP manage the QP states is can call rdma_reject 
after getting the REP instead of rdma_accept and this will do what you need.

The idea behind the orig CM***A*** design was to have MAX abstraction in 
the price of MIN (ZERO) impact on the ULP. We were thinking here that 
after getting the REP its fine to send REP and deliver up ESTABLISHED.

You are suggesting a design change in the CMA which would effect also 
the current CMA ULP consumers: iSER, RDS, NFSoRDMA and Lustre.

Or.